Show HN: HNtropy - interesting story pairings from across the Hacker News corpus
I shared earlier versions of this project using traditional ranks (points, comments, discussion size, and title/topic similarity). They worked, but after posting them I realized most top results were familiar clusters, and a lot of that territory is already covered by HN Algolia.
So I rebuilt the ranking around a different signal: explicit “related:” style cross-links inside HN comments plus the broader HN backlink graph (when discussion A links to discussion B). That graph is the core signal. From there, I score pairings by evidence strength and then favor links with lower lexical overlap, so we surface semantically connected stories that don’t look identical in title wording.
What I learned: big threads are often strong on attention, but weaker on novelty. Backlink + “related:” intent is a better proxy for meaningful cross-thread connection. If you want standard topic clustering, Algolia is perfect; this is for high-entropy pairings discovered from HN’s own linking behavior.
Check it out:
No comments yet.